Why everyone is talking about generative AI, not just experts

The ability of machines to generate images and text has advanced tremendously over the past decade. As is often the case with innovation, progress is not linear but leaps and bounds, surprising and delighting researchers and users. 2022 was a symbolic year for innovation in generative AI, based on the advent of diffusion techniques for image generation and the emergence of increasingly large-scale transformers for text generation.

Although it’s made a huge leap forward for the entire natural language processing (NLP) industry, there are three reasons why generative AI models first captured the public’s excitement. A language that AI can do for the time being.

What’s Behind the Generative AI Excitement?

The most obvious reason is that they fall into a highly intuitive class of AI systems. These models are not used to create high-dimensional vectors or uninterpretable code, but rather natural-looking images and fluent, coherent text that anyone can see and understand. used for People outside of machine learning don’t need specific expertise to judge how natural or fluent a system is. As such, this part of AI research seems much more approachable than other (perhaps equally important) areas.

Second, there is a direct relationship between generations and how we assess intelligence. generation answer to the ability to discriminate Answer by choosing the correct answer. I believe that having students explain things in their own words helps them understand the topic better. Eliminate the possibility of students simply guessing or memorizing the correct answer.

This is how we feel when artificial systems produce natural images and coherent prose. forced We can compare it to similar human knowledge and understanding, but whether this is overly lenient on the actual capabilities of man-made systems is an open question in the research community. It is clear from , that the model’s ability to generate novel yet plausible images and text depends on a rich internal representation of the underlying domain (e.g., the task at hand, what the image or text is “about”). type). ) are included in these models.

Moreover, these representations are useful in a broader domain than just generation for generation. In short, generative models were the first models to catch the public eye, but many more valuable use cases will emerge in the future.

one to another

Third, modern generative models demonstrate the ability to conditionally generate. Instead of sampling existing images or snippets of text, you have the ability to create text, videos, images, or other modalities. conditional to something else, such as partial text or an image.

To understand why this is important, we need to look at most human activities that involve producing something in dependence on something else. To name a few:

• Writing an essay is creating a text based on questions/topics and knowledge and views contained in our own experiences and books, papers and other documents.
• Having a conversation is generating knowledge about our world, an understanding of the pragmatics that the situation requires, and a response that is contingent on what has been said up to that point in the conversation.
• Drawing an architectural plan involves drawing an image based on architectural and structural engineering principles, sketches or photographs of the terrain and its topology/surroundings, and our knowledge of the (often under-specified) requirements provided by the client. to generate.

Most intelligent behavior follows this pattern of generation. something based on Others as a context. The fact that man-made systems now have this ability means that our jobs are likely to become more automated, or at least more symbiotic relationships between humans and computers. You can already see this in new tools like CodeWhisperer that help humans code, and new tools like Jasper that help you write marketing copy.

Today, we have systems that can create text, images, or videos based on other information you feed. That means we can apply these generations to similar problems and processes that once required human expertise. This will lead to more automation, or a more symbiotic form of support between human and artificial systems, with practical and economic consequences.

new basic tools

For the rest of 2023, the big question is what all this progress really means in terms of potential applications and utilities. It’s very exciting to be in the industry because we’re building the fundamental tools for building intelligent systems and processes, making them as intuitive and applicable as possible, and putting them in the hands of the widest class. It’s time. Developers, builders and innovators are possible. This is what drives my team and our mission to help computers communicate better with us and communicate using language.

Human intelligence is more than just the processes this technology enables, but combined with the limitless ability humans have to constantly innovate, aided by new tools and technologies, the innovation we see in 2023 will change. I have no doubt that it will. How to use your computer in destructive and wonderful ways.

Ed Grafenstedt teeth Head of Machine Learning at Cohere.